Just one more quick opinion - I do BGV's with Melodyne/Autotune/whatever all the time, but it depends on the kind of BGV's. Of course you can't use it to generate a full secondary melodic line with it's own words and such, but for just a tight "follow on the 3rds" type harmony that will go on a 2nd verse or in spots to add emphasis it's just a matter of time and money. I've already recorded 5 or 10 takes of the lead part and comped and corrected them, so at that point I have two choices:
 
1 - bring the lead vox singer back in to do the spot harms and repeat the recording/comping/correction process, or
 
2 - take the existing lead part and do it in a few minutes without having to even get a mic out, let alone a de-esser and Vocal Rider (and comp, and trim, and de-breath, and adjust timing, and, and, and...).
 
For most of the bands I work with, particularly for the singer-songwriter types that are just used to playing and singing themselves, it's just a no-brainer, both in terms of my time and their pocket book. Honestly, pretty much everyone I know does it that way...it's pretty ubiquitous at this point.
 
I mean, unlimited time and a bottomless wallet? Sure, I'll have 'em sing every part and do the work, but until that happy day arrives I'm using Melodyne, and honestly, Sonar's ARA implementation is one of the big reasons I stay with Sonar.
 
Dean

I do it backwards and wrong (I realize this will probably come as no surprise to anyone here ). I use Melodyne to create a harmony from the lead vocal, and play around with the notes until the harmony sounds good with the lead. Then I learn the harmony line and sing it with my voice. When the vocal is done, I delete the Melodyne part.
 
Because it's easy to compare both the Melodyne and "real" vocals, most of the time the real vocals sound better. However, on rare occasions the Melodyne harmony's timbre can be the right fit. Also, running the voice through Melodyne Editor to change formant but nothing else can give some interesting variety to background vocals.
 
My latest BGV trick is VocalSync. I'm getting consistently good results with it, but find it's at its best when you move vocals closer together, not always right on top of each other.
 
Finally, I try to have BGVs that do some level of call and response and "hole filling," not just singing the lead vocal but with different notes.

I write out parts for the vocalist(s)—yes, on music paper!—and I also have a little electronic keyboard I bought at Toys R Us for $49 for cueing or for demonstrating the notes.  That's easier than changing which track is echoing, having to unfreeze a piano track, etc.
 
Off topic, but I plan to use some of the crummy sounds from that little keyboard on my next album (tentatively titled "Kellis Park").  There are some weird Lucy in the Sky sounds I really like.

Having access to the singers is a valid point.
 
I did make a pass at this once to help another user (singer) see how capable Melodyne was with doing such. IIRC the biggest gain/variation was had by lopping consonants out of the track and just working with the vowel sounds. Even just using Melodyne to lop off consonants allowed for much more extreme vowel transposition than I expected (using several different tools/methods).

I also do my own harmonies and background vocals.   But I also use Melodyne and VocalSync (great addition).   The only times I've ever "created" harmony parts using Melodyne (or VVocal before) is when I simply cannot sing them. With a little care and some blending, it works.  But like others I don't go for perfection, I go for what I think sounds best in context.